I disagree that's it's solved in practice for all users. I agree it's solved for many well established workloads. Researchers are the people most in need of ra

I disagree that's it's solved in practice for all users. I agree it's solved for many well established workloads. Researchers are the people most in need of rapid experimental iteration vs one off training. But during research the data requirements may not be pinned down well enough to pre-process everything ahead of time or repeatedly doing so is it's own bottleneck. Out of core loading may still be a bottleneck if IO is slow, samples are big or necessary pre-processing can't be pushed to GPU.